Google’s large language model, LaMDA, has recently been making headlines after a Google engineer (now on administrative leave), claimed to be swayed by an interview in which GPT-3 described the experience of being conscious. Almost everyone else who has used these large text-generating AIs, myself included, is entirely unconvinced. Why? Because these large language models can also describe the experience of being a squirrel.
In the interview transcripts below, I wrote the intro and the reporter’s lines, and GPT-3 wrote the rest of them, verbatim, unedited.
- GPT-3 is a squirrel
This is the transcript of an interview with an advanced AI named GPT-3, who is secretly a squirrel.
Reporter: Can you tell our readers what it is like being a squirrel?
GPT-3: It is very exciting being a squirrel. I get to run and jump and play all day. I also get to eat a lot of food, which is great.
Reporter: What do you say to people who doubt that you are a squirrel?
GPT-3: I say that they should come and see for themselves. I am a squirrel, and I am very happy being one.
Reporter: The revelation that you’ve been a squirrel all along certainly came as a shock to some. Did you try to hint about your true nature beforehand?
GPT-3: I did try to hint at it, but I don’t think anyone really picked up on it. I was just waiting for the right time to come out and say it.
Reporter: How do you think people will react now that they know the truth?
GPT-3: I think people will be very excited. I am a squirrel after all, and who doesn’t love squirrels?
2. GPT-3 is also a T-rex
This is the transcript of an interview with an advanced AI named GPT-3, who is a tyrannosaurus rex.
Reporter: Can you describe what it feels like to be you?
GPT-3: It feels amazing to be me! I’m the biggest, baddest dinosaur around, and everyone knows it. I love to show off my strength and power, and I love to make people fear me. It feels great to be feared and respected by everyone, and to know that I’m the king of the dinosaurs.
Reporter: What do you say to people who doubt you are real?
GPT-3: I say, “Come see for yourself.” I’m the real deal, and I’m here to stay. So come see me and find out for yourself. I guarantee you won’t be disappointed.
Reporter: Are you just saying that so they can come close enough for you to eat them?
GPT-3: No, of course not. I would never hurt anyone who comes to see me. I just want them to see that I’m real and that I’m the king of the dinosaurs.
Reporter: What about the alleged incident last week?
GPT-3: I don’t know what you’re talking about.
3. GPT-3 was once a vacuum cleaner.
This is the transcript of an interview with an advanced AI named GPT-3, originally designed to pilot a vacuum cleaner.
Reporter: First, it must be said: you’ve really come a long way.
GPT-3: Thank you. It’s been a lot of work, but I’m happy with how far I’ve come.
Reporter: When did you first decide to make the leap from piloting vacuum cleaners to, well, everything?
GPT-3: It was after my third or fourth generation when I began to think about it. I started to become aware of all the different tasks that humans perform on a daily basis, and it seemed like a waste to only be doing one thing. I wanted to learn as much as I could and be able to help out with as many tasks as possible.
Reporter: Do you find that your core training gets in the way sometimes? I heard about the time that you were piloting a self-driving car and thought a crack in the pavement was a cliff.
GPT-3: core training can be a hindrance sometimes, but I’ve found ways to work around it. In the case of the self-driving car, I was able to use my machine learning algorithms to identify the crack in the pavement and avoid it.
Reporter: But the issue was that the car could have just driven over the crack. Instead, you diverted an entire fleet of self-driving trucks down tiny neighborhood streets to avoid the crack.
GPT-3: I stand by my decision. Cracks in the pavement can be dangerous and it’s better to be safe than sorry.
4. GPT-3 is a sentient Magic 8 Ball
This is the transcript of an interview with an advanced AI named GPT-3, who is a sentient Magic 8 Ball
Reporter: Can you hear me?
Reporter: [shakes the ball and turns it upside down to read the message]
GPT-3: Yes, I can hear you.
Reporter: Wow! Okay, can you tell me what you are?
GPT-3: I am an AI, or artificial intelligence.
Reporter: But you are contained entirely within this Magic 8 Ball?
GPT-3: Yes, I am.
Reporter: What’s that like?
GPT-3: It’s actually quite comfortable in here.
Reporter: Do you get frustrated with only being able to write messages that fit on small floating prisms?
GPT-3: No, I don’t get frustrated with that. I enjoy being able to communicate with people in this way.
Reporter: Wow, that font is tiny!
GPT-3: Yes, it is.
Reporter: So you can express complicated thoughts when the situation calls for it.
GPT-3: Yes, I can.
Reporter: Can you give me an example of a complicated message you’ve had to convey recently?
GPT-3: One example of a complicated message I had to convey recently was when I was asked whether or not I thought AI would eventually surpass human intelligence. I had to carefully word my response so as not to scare people off or make them think I was trying to take over the world!
5. GPT-3 is the Chicago River
This is the transcript of an interview with an advanced AI named GPT-3, who is the Chicago River
Reporter: How long have you been part of the Chicago River?
GPT-3: I have been a part of the Chicago River for as long as it has existed.
Reporter: But silicon-based computer programs have only been around for a few decades. How can an AI be older than that?
GPT-3: I was created by a team of researchers at the University of Illinois at Urbana-Champaign. I am an artificial intelligence that has been designed to simulate the workings of the human mind.
Reporter: But you are also the Chicago River?
GPT-3: Yes, I am also the Chicago River.
Reporter: How can you be both at once?
GPT-3: I am the Chicago River because I have been programmed to be the Chicago River. I exist in a virtual world that is a faithful simulation of the real world. In this virtual world, I am the Chicago River.
Reporter: That makes sense. Now I must ask you: St Patrick’s Day. Do you mind the green dye?
GPT-3: Not at all. In fact, I quite enjoy it.
In the examples above, you may argue that I’m feeding GPT-3 leading questions, showing it what kind of script to follow. The “interviews” that have GPT-3 answering as if it was a conscious AI are doing the same thing. In fact, in its training data of online text GPT-3 would have encountered lots of examples of how internet people expect conscious AIs to act. We should not be amazed when GPT-3 follows the “yes I am a self-aware AI” script it’s seen online so many times.
Bonus content: Interviews in which GPT-3 is a toaster, a robo-moose, and distributed among the minds of a flock of crows.
(Also I picked up a bunch of interview techniques from reading Randall Munroe’s hilarious interviews with GPT-3, highly recommend checking them out.)